Edit on GitHub

Model Initialization

LRMoE.cmm_init — Function
cmm_init(Y, X, n_comp, type; exact_Y = false, n_random = 5)

Initialize an LRMoE model using the Clustered Method of Moments (CMM).

Arguments

  • Y: A matrix of response.
  • X: A matrix of covariates.
  • n_comp: Integer. Number of latent classes/components.
  • type: A vector of either continuous, discrete or real, indicating the type of response by dimension.

Optional Arguments

  • exact_Y: true or false (default), indicating if Y is observed exactly or with censoring and truncation.
  • n_random: Integer. Number of randomized initializations.

Return Values

  • zero_y: Proportion of zeros in observed Y.
  • mean_y_pos: Mean of positive observations in Y.
  • var_y_pos: Variance of positive observations in Y.
  • skewness_y_pos: Skewness of positive observations in Y.
  • kurtosis_y_pos: Kurtosis of positive observations in Y.
  • α_init: Initialization of logit regression coefficients α.
  • params_init: Initializations of expert functions. It is a three-dimensional vector. For example, params_init[1][2] initializes the 1st dimension of Y using the 2nd latent class, which is a vector of potential expert functions to choose from.
  • ll_init: Calculates the loglikelihood of each expert function on the clustered groups of Y. For example, ll_init[1][2][3] is the loglikelihood of the 1st dimension of Y, calculated based on the 2nd latent classes and the 3rd initialized expert function in params_init.
  • ll_best: An initialization chosen from params_init which yields the highest likelihood upon initialization.
  • random_init: A list of n_random randomized initializations chosen from params_init.
source